Claude Skills by tomzx

Proactively discover latent bugs in an existing codebase (correctness, error handling, null and boundary handling, async, concurrency, state, resource management), keep only high-confidence ones, verify each where a seam exists, and use ghx to discard anything already tracked by a GitHub issue, searching for another candidate until nothing useful remains. Use when the user says /identify-bugs, "find bugs", "hunt for bugs", "look for latent bugs", "bug sweep", "what is broken", or wants a peri...

Discover concrete gaps and improvements in an existing codebase, keep only the high-confidence ones, and use ghx to discard anything already tracked by a GitHub issue, searching for another candidate until nothing useful remains. Works bottom-up from the code (correctness, tests, maintainability, performance, dependencies, DX) rather than from the backlog. Use when the user says /identify-codebase-improvements, "find improvements", "what should we improve", "codebase gaps", "tech debt hunt", ...

Analyze the existing software surface and surrounding signals (code, issues, recent PRs, goals, roadmap) to generate and rank new feature opportunities that would add value. The bottom-up discovery counterpart to create-roadmap and create-goals, and the inverse of create-needs-assessment (which validates a given idea rather than proposing one). Use when the user says /identify-feature-opportunities, "what should we build next", "find feature ideas", "product opportunities", "feature gaps", "w...

Identify workflows the user performs that are not yet encoded as skills in the agents repository, and prompt to create or backlog them. Drives a self-reinforcing improvement loop.

Reflect on the current session and identify what would have been needed to run it fully autonomously end to end, producing a concrete readiness assessment and a prioritized gap list. Use when the user asks /improve-autonomy, "what would it take to run this autonomously?", or wants to assess autonomy readiness.

Make a bounded set of safe, verified improvements to a codebase each run (remove dead code, add missing types, auto-fix lint, drop unused deps, patch dependencies). Designed to run weekly as a routine. Use when the user says /improve-codebase, "improve the codebase", "weekly cleanup", "tend the repo", or wants incremental hands-off improvements that pass the project's own tests and open a reviewable PR.

Mine past AI coding sessions from the agentsview SQLite archive and produce prioritized recommendations for improving future sessions (recurring friction, corrections, repeated questions, manual loops, missing skills). Use when the user says /improve-sessions, "improve future sessions", "what should I learn from my sessions", "review my session history for improvements", "why do my sessions keep going off the rails", or wants recommendations derived from session history rather than the curren...

Inspect a single skill in the library, reason about high-value low-risk improvements, and apply them directly. Used by the Skill Refresher workflow and invocable directly as /improve-skill <skill-name>.

Bootstrap a developer trust profile by scanning the last N PRs they authored (merged and rejected), determining outcome from actual review results.

Bootstrap the .sdlc/ directory structure in a project, creating subdirectories and populating templates.

Builds theme-organized Slack channel knowledge from temporal monthly memory files. Use when merging notes across months for a given channel, deduplicating by topic, or maintaining the non-temporal “by theme” KB for that channel.

Add relevant labels to a GitHub issue based on its description, and recommend new labels if needed.

Collect the code review feedback you received on your own pull requests and distill it into durable, cited rules to consult before design and implementation work so the same mistakes stop recurring. Use when the user says /learn-from-code-reviews, "learn from my code reviews", "learn from my PR reviews", "what do reviewers keep telling me", "turn review feedback into rules", "why do I keep getting the same review comments", "review my feedback history", or wants a review-rules checklist to ap...

Check that a pull request is approved and CI is passing, then merge it and clean up the branch.

Check SLOs/SLIs, review error rates, latency, and throughput for deployed features.

Bootstrap SDLC, configure automated issue triaging, and apply standardization skills to bring a repository in line with the practices encoded in this skill library. Run once per repository to establish baseline structure, health assessment, and continuous triage.

Run the full product development lifecycle loop, from discovery through launch and measurement, with proceed/pivot/kill gates at every phase.

Map the competitive landscape and positioning gaps for a candidate problem space. Part of the PDLC Discover phase.

Trace shipped product outcomes back to the original problem and metrics, PDLC's equivalent of backpropagation. Catches drift where the spec was built faithfully but stopped being the right thing.

Produce an exec summary, status update, or stakeholder map for a product initiative or the whole portfolio. The communication layer of the PDLC.

Sequence initiatives across Now/Next/Later horizons, aligned to goals and outcomes. Part of the PDLC Strategy phase.

Produce the value proposition, messaging house, and audience-specific narratives for a launch. A consumer of the validated vision and pricing. Part of the PDLC Launch phase.

Produce the acceptance contract — testable criteria that are the seam between PDLC and SDLC. PDLC owns what/why; SDLC owns how.

Define the product vision, target customer, and positioning for a validated opportunity. Start of the PDLC Strategy phase; writes the product-level vision anchor.

Design the cheapest decisive test for an initiative's riskiest assumptions — fake-door, concierge, prototype, or survey. Part of the PDLC Validate phase.

Synthesize customer interviews, jobs-to-be-done, and signals into structured problem statements. The divergent start of the PDLC Discover phase.

Produce the sales playbook, support runbook, and internal training needed so teams can sell, support, and explain the launch. Part of the PDLC Launch phase.

Convert discovery inputs into a scored opportunity-solution tree. Produces the artifact the first PDLC gate evaluates.

Bootstrap the .pdlc/ directory structure in a project, creating subdirectories and populating templates.

Explicitly stop a product initiative — reallocate roadmap capacity, record the sunset-or-pivot rationale, and notify stakeholders. The inverse of prioritize.

Record a product decision, including the PDLC proceed/pivot/kill gate verdict at each phase boundary. One reusable decision record, invoked at every gate.

Assess health across multiple product initiatives and the overall portfolio — balance, bet distribution, and capacity allocation.

List the hypotheses, beliefs, and risks that must be true for an opportunity to succeed, ranked by leverage. Start of the PDLC Validate phase.

Display a progress dashboard for product initiatives from .pdlc/ directory data, without modifying any artifacts.

Plan a product launch — readiness checklist, timeline, channels, and success criteria. Start of the PDLC Launch phase.

Rank a backlog of initiatives or features using a transparent scoring framework (RICE / WSJF). End of the PDLC Strategy phase.

Create wireframes or a clickable prototype and run a lightweight usability test, recording findings. Part of the PDLC Define phase.

Size the market, map trends and segments, and assess timing for a candidate problem space. Part of the PDLC Discover phase.

Read the dashboards, compare actuals against goals, surface anomalies, and produce a health report. Part of the PDLC Measure phase.

Execute an experiment plan, collect data, and record a proceed/kill verdict against the pre-set threshold. End of the PDLC Validate phase.

Capture what to keep, stop, and start after an initiative, sprint, or kill. Closes the PDLC cycle by turning experience into durable knowledge.

Define OKRs, success metrics, and guardrail metrics for a product or initiative. Part of the PDLC Strategy phase.

Define the pricing model, packaging, and tiers for an initiative. A consumer of the validated value and willingness-to-pay evidence. Part of the PDLC Launch phase.

Specify the events, funnels, dashboards, and guardrails needed to measure an initiative. Guardrail-first by design. Start of the PDLC Measure phase.

Plan the end-of-life for a product or feature — migration, communications, and timeline. A first-class PDLC phase, triggered when an initiative is net-negative.

Reconcile the .pdlc/ directory with current product reality — create the structure if absent, detect drift, and ensure the anchor exists. The PDLC counterpart of sync-sdlc.

Aggregate qualitative signal — support, sales, NPS, reviews — and combine it with the metrics read. Produces the artifact the iterate/sunset gate evaluates. Part of the PDLC Measure phase.

Write a product requirements document — problem, outcomes, success criteria, and non-goals. Start of the PDLC Define phase.

Post a message to a Slack channel or as a thread reply. By default writes the message to a temp file and asks the user to confirm before sending; pass --yes to send immediately without prompting. Use when the user says /post-slack-message, "post to Slack", "send a Slack message", "reply in a Slack thread", or another skill delegates its Slack posting here.

Send PR review comments to GitHub by file and line number using the pr-comment script.